I want to generate a following array a:
nv = np.random.randint(3, 10+1, size=(1000000,))
a = np.concatenate([np.arange(1,i+1) for i in nv])
Thus, the output would be something like -
[0, 1, 2, 3, 0, 1, 2, 3, 4, 0, 1, 2, 0, 1, 2, 3, 4, 5, 0, ...]
Does there exist any better way to do it?
Here's a vectorized approach using cumulative summation -
def ranges(nv, start = 1):
shifts = nv.cumsum()
id_arr = np.ones(shifts[-1], dtype=int)
id_arr[shifts[:-1]] = -nv[:-1]+1
id_arr[0] = start # Skip if we know the start of ranges is 1 already
return id_arr.cumsum()
Sample runs -
In [23]: nv
Out[23]: array([3, 2, 5, 7])
In [24]: ranges(nv, start=0)
Out[24]: array([0, 1, 2, 0, 1, 0, 1, 2, 3, 4, 0, 1, 2, 3, 4, 5, 6])
In [25]: ranges(nv, start=1)
Out[25]: array([1, 2, 3, 1, 2, 1, 2, 3, 4, 5, 1, 2, 3, 4, 5, 6, 7])
Runtime test -
In [62]: nv = np.random.randint(3, 10+1, size=(100000,))
In [63]: %timeit your_func(nv) # @MSeifert's solution
10 loops, best of 3: 129 ms per loop
In [64]: %timeit ranges(nv)
100 loops, best of 3: 5.54 ms per loop
Instead of doing this with numpy methods you could use normal python ranges and just convert the result to an array:
from itertools import chain
import numpy as np
def your_func(nv):
ranges = (range(1, i+1) for i in nv)
flattened = list(chain.from_iterable(ranges))
return np.array(flattened)
This doesn't need to utilize hard to understand numpy slicings and constructs. To show a sample case:
import random
>>> nv = [random.randint(1, 10) for _ in range(5)]
>>> print(nv)
[4, 2, 10, 5, 3]
>>> print(your_func(nv))
[ 1 2 3 4 1 2 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 1 2 3]
Why two steps?
a = np.concatenate([np.arange(0,np.random.randint(3,11)) for i in range(1000000)])
np.rangeshould benp.arange, I believe.np.r_[:4, :5, :3, :6]is a nice compact way of generating such an array 'by-hand'. It doesn't offer any advantages if starting with anvarray.